Multivariate creative testing without a data team is easier than most marketers assume. Run a fractional design: vary two elements across a small set, treat the output as directional, and use the winner to brief the next batch. You do not need to test every combination to learn which element moves the result.
What does a fractional multivariate test look like?
A fractional design tests a subset of possible combinations rather than every cell in a full factorial. The most practical version for creative is a 2x3: two variants of one element against three variants of a second element. That gives six ads.
Say you want to isolate hook and CTA. Hook has two variants: a question hook and a founder quote hook. CTA has three variants: "Shop now", "Get 20% off", and "Learn more".
| Variation | Hook | CTA |
|---|---|---|
| 1 | Question | Shop now |
| 2 | Question | Get 20% off |
| 3 | Question | Learn more |
| 4 | Founder quote | Shop now |
| 5 | Founder quote | Get 20% off |
| 6 | Founder quote | Learn more |
This is a full factorial for those two elements, but fractional against the wider creative space. Script, shot selection, voiceover, captions, aspect ratio and offer all normally vary in a full programme. In a fractional test you hold everything except the two chosen elements constant.
Full factorial versus fractional: the cost difference
A full factorial tests every combination of every element you care about. Three hooks times three CTAs times two proof styles is 18 cells. Add a fourth element and you are at 27 or more. A fractional 2x3 stays at six.
| Design | What you vary | Ads to produce | Cost in Genyad credits | What you learn |
|---|---|---|---|---|
| Full factorial 3x3x3 | Hook, CTA, proof | 27 | 27 credits | Main effects and interactions, if volume allows |
| Full factorial 4 factors at 2 levels | Hook, CTA, proof, caption style | 16 | 16 credits | Broader coverage, still deep on no single interaction |
| Fractional 2x3 | Hook, CTA | 6 | 6 credits | Directional read on which element moves performance |
The cost difference is not only production. Each cell still needs enough impressions for a stable read. A 27-cell test can require far more media budget, more review time and more patience than a six-cell test. On Genyad, one video ad variation costs one credit, so a six-cell test costs six credits. Growth is €99 for 65 credits, enough for weekly creative tests without paying an editor to recut the same timeline.
A simple creative testing calculator can tell you whether your expected volume will support six cells before you waste budget. Most accounts see that six cells is the safer starting point.
What you give up when you skip cells
Fractional tests abandon interaction reads. If the founder quote hook works only with "Get 20% off" and not with "Shop now", a 2x3 may show mixed hook performance and hide that pairing. You get a direction, CTA matters more than hook, but not a specific winning pair.
You also give up segmentation. A fractional test will not tell you whether a question hook works for women aged 25 to 34 on Reels but not on feed. That is left to a follow-up test.
Confounding is the third trade-off. If one CTA variant gets paired with a hook that happens to have stronger footage, the read can be noisy. Accept directional answers as exactly that: enough to decide what to test next, not enough to declare a universal law.
This is why creative testing should be sequential. Run a 2x3, find the element with the bigger signal, then test the next element against the winner. You do not need one giant factorial.
When to hand it to DCO instead
DCO is the better call when you have enough conversion volume and audience signal to let the platform optimise combinations for you. Meta and Google can mix elements per user, shift budget to the combinations that convert, and report at the cell level. That is the promise of dynamic creative optimisation, and it works when the machine has enough data.
Manual fractional testing is better when volume is thin, when you need control over scripts, or when you want to learn which elements are worth feeding into DCO later. The two approaches are not rivals. A 2x3 test can tell you that CTA matters more than hook. Then you hand a larger DCO set to the platform with more CTA variants and fewer hook variants.
Genyad does not run DCO. We generate variations from your existing footage library, but we do not score predicted performance and we do not publish directly to Meta or TikTok. If your account has enough volume for platform DCO, use that for real-time optimisation and use Genyad to build the variations from footage you already own.
How to run a 2x3 fractional test with Genyad
Upload your footage once. Genyad transcribes and tags every clip. You do not need an editor to create the six variations. Ask for two hook versions and three CTA versions across the same library. Each variation is a fresh script, shot selection, voiceover, caption set and export, not a recut of the same timeline.
The six variations cost six credits total. Keep everything else constant: same offer, same product, same landing page, same aspect ratio. Run all six in the same campaign or ad set, with the same budget per cell if possible.
Then read the result directionally. If all three CTA variants outperform both hook variants on conversion, the CTA is the stronger lever. If the question hook wins across all CTAs, hook matters more. Do not overread a single cell.
Our 2026 fatigue benchmark is a useful pressure test here. It synthesises published platform and agency figures and puts most creative as effectively dead within three weeks. That means throughput, not one perfect test, is the bottleneck. A six-cell fractional test you can ship this week beats a 27-cell factorial you never launch.
Frequently asked questions
What is the difference between multivariate and A/B testing?
A/B testing changes one element at a time. Multivariate testing changes more than one element in the same set. A fractional multivariate test is the middle ground: two elements, a small grid such as 2x3, and a directional read rather than a full interaction analysis.
How many variations do I need for a fractional test?
A 2x3 gives six variations. An eight-cell 2x2x2 works if you need to test three elements. More than eight often becomes expensive before you know the direction, especially if each cell needs enough impressions to harden the read.
Does Genyad publish the winning variation to Meta?
No. Genyad does not publish directly to Meta or TikTok. You export the video and upload it in Ads Manager. Genyad also does not provide predicted performance scores.
When should I use DCO instead of a manual fractional test?
Use DCO when you have enough conversion volume and want the platform to combine elements per user and shift budget automatically. Keep manual fractional testing when volume is thin, when you need script control, or when you want to learn which elements to feed into DCO later.
Can I run a fractional test with only five credits?
Yes. The free plan gives five video ad variations, so you can run a five-cell test or a smaller 2x2 plus one extra variation. Starter is €29 for 15 credits, enough for two six-cell tests with three spare variations.